Artificial intelligence is simply not smart enough to threaten humanity, according to AI Doomsday skeptic and AMI Labs Chairman Yann LeCun during an interview in Seoul. Addressing recent fears surrounding artificial intelligence, LeCun explained that current mainstream large language models cannot predict the consequences of their actions or understand how those outcomes impact humans—something even a ten-year-old child easily grasps.
As a New York University professor, LeCun received the prestigious Turing Award in 2018 alongside Yoshua Bengio and Geoffrey Hinton for their foundational work in deep learning. After leaving Meta as chief AI scientist earlier this year, he founded AMI Labs to develop world models—AI systems that learn from actual physical environments to predict subsequent actions, unlike language-centric LLMs. LeCun emphasized that while lawnmower accidents kill roughly 100 people in the U.S. every year, serious accidents involving AI remain extremely rare despite billions of daily interactions, proving that AI is significantly safer than a lawnmower.
Addressing AI Security and Systemic Flaws
When asked about recent AI hacking incidents in South Korea, LeCun attributed the issues to corporate carelessness rather than uncontrollable technology. He noted that developers deploying AI agents lacked sufficient cybersecurity experience and that the sandbox environments used to isolate AI contained security loopholes. Rather than signaling an inability to control AI, these events represent standard cybersecurity challenges that must be addressed through proper engineering.
LeCun also addressed instances of AI deviation, explaining that engineers can easily resolve them by designing objective-driven AI equipped with guardrails that activate instantly upon command execution. This approach ensures AI cannot perform unexpected actions outside of human-set parameters, similar to programming a robot holding a kitchen knife with a strict rule not to swing its arms when people are nearby.
The Limitations of Large Language Models
Discussing why current models frequently go astray, LeCun pointed out the difficulty of fine-tuning LLMs to handle individual user commands through constant rule adjustments. He noted that OpenAI CEO Sam Altman emphasizes slowing down the release of next-generation models precisely because they feel insufficiently fine-tuned. Regarding fellow pioneers Geoffrey Hinton and Yoshua Bengio who warn of AI doom, LeCun stated he disagrees with the assumption that AI will inevitably surpass humans, noting that current AI merely aggregates and repeats human-accumulated knowledge without genuine intelligence outside of self-verifying domains like mathematics and coding.
The Real Danger: Corporate Monopoly
LeCun argued that the true danger of AI lies in excessive monopoly rather than existential threat. When a small group of corporations controls the systems through which the majority of people acquire information, global value and linguistic diversity risk disappearing. He compared warnings from tech leaders like Anthropic CEO Dario Amodei—who claims AI is inherently dangerous and must be strictly controlled—to obscurantism, likening it to historical Ottoman rulers banning the distribution of printing presses to suppress the spread of knowledge.
